Problem
Instruction tests under mlir/test/Target/DXSA/ that use the inline hex form (mlir-translate --import-dxsa-hex %s) carry a .mlir extension,
but their payload is tokenized DWORDs plus // RUN/// CHECK lines, not MLIR. The .mlir extension makes editors apply MLIR syntax highlighting to hex input, which is misleading. (Raised by @hvdijk in #167.)
Proposal
Rename all inline-hex tests to .test, LLVM's conventional suffix for RUN+CHECK driver tests whose input is not a specific source language.
Do this as a single sweep so the convention stays consistent across the dialect, rather than per-PR.
Notes
The inline hex tokens are a temporary solution until we have a proper binary writer (the serialization / round-trip path we don't have yet). Once that exists the way we feed inputs to these tests may change, so we may want to revisit the test format then.
Problem
Instruction tests under
mlir/test/Target/DXSA/that use the inline hex form (mlir-translate --import-dxsa-hex %s) carry a.mlirextension,but their payload is tokenized DWORDs plus
// RUN/// CHECKlines, not MLIR. The.mlirextension makes editors apply MLIR syntax highlighting to hex input, which is misleading. (Raised by @hvdijk in #167.)Proposal
Rename all inline-hex tests to
.test, LLVM's conventional suffix for RUN+CHECK driver tests whose input is not a specific source language.Do this as a single sweep so the convention stays consistent across the dialect, rather than per-PR.
Notes
The inline hex tokens are a temporary solution until we have a proper binary writer (the serialization / round-trip path we don't have yet). Once that exists the way we feed inputs to these tests may change, so we may want to revisit the test format then.